Zillow may have paid $1 million to the winners of its Zillow Prize, a competition aimed at improving its long-running Zestimate, but it looks like some of the findings from the winning team are paying off for the home-valuation tool.
The Seattle-based real-estate technology company announced Thursday that the new Zestimate is the “most sophisticated and accurate” version since it was launched in 2006. To get there, the company credits the use of real-time data as well as computer vision that was trained on millions of photographs of homes on Zillow.
Zillow said that for homes listed for sale, the Zestimate error rate is now less than 2 percent of the home’s eventual sale price.
Ideas from the two-year Zillow Prize, a data-science competition that attracted more than 3,800 teams representing 91 countries, have been incorporated into the Zestimate. Sophisticated machine learning techniques now search photos for special elements and details of a home and neural networks crunch the pixels and adjust value accordingly.
“The new Zestimate was inspired by the way the human brain interprets scenes, objects, and images,” Stan Humphries, chief analytics officer and creator of the Zestimate, said in a news release. “It’s a big leap forward, because it means the Zestimate can now understand not just a home’s facts and figures, but its quality and curb appeal. By training neural networks using the millions of home photos on Zillow, the Zestimate now values the features people are most proud of — like new granite countertops, a remodeled bathroom, or a meticulously landscaped backyard.”
Real-time data from for-sale homes, including list price and how long a home has been on the market are also new keys to the tool.
Zillow consistently refers to the Zestimate as just one data point that consumers have access to when considering buying or selling a home — along with information such as recent home sales and guidance from real estate professionals. Zillow explicitly points out that the Zestimate does not constitute an appraisal and is what it sounds like, an estimate. Thirteen years ago, it marked the first time that homeowners gained access to estimated home values — data that was previously only available to real estate agents, appraisers and mortgage lenders.
The Zestimate is not without critics.
A group of homeowners in Illinois sued Zillow in 2017, alleging that the Zestimate tool is often inaccurate and difficult to get changed, and that Zillow markets it as roughly equivalent to an appraisal. The homeowners argued that the tool undervalued their homes and made it harder for them to sell. A federal appeals court sided with Zillow in the lawsuit this past February.
Jordan Meyer of the United States, Chahhou Mohamed of Morocco, and Nima Shahbazi of Canada made up a team that won the Zillow Prize, which former CEO Spencer Rascoff called one of the largest computer science contests in the history of technology.
So where does the Zestimate go from here? Zillow has some ideas, with the understanding that many some things just can’t be predicted.
“One area of research we’re exploring now is how we can create even more hyper-local and hyper-specific models capable of discerning how different housing markets value certain home features,” a Zillow spokesperson told GeekWire. “We’ll likely never be totally perfect, since so much of the home buying and selling experience can be really personal and emotional. There are variables that a machine could never predict, like how much someone will pay for a house that reminds them of their childhood home and the good memories they had there.
“But we’re always looking for new ways to improve the Zestimate.”